Cell fate specification in the developing Drosophila retina

Although the complex nature of developing neural tissues often makes them difficult to study, most share certain features common to all developing tissues. In particular, the specification of cell fate within most developing epithelia and neuroepithclia so far examined makes use of inductive interactions between cells. The importance of communication between neighboring cells is beginning to be appreciated, and elucidating the molecular mechanisms by which one cell can direct the fate of its immediate neighbors presents an exciting challenge. In this review. I will focus on recent progress in studies concerning one such developing neuroepithelium, the retina of the fruitfly Drosophila melanogasler. The fly retina represents a simple micro-nervous system, and the genetic and molecular tools developed for Drosophila have proven particularly powerful for its study. In addition, development of the fly retina has been described in morphological detail (Ready et al., 1976: Tomlinson. 1985; Cagan and Ready. 1989b; Wolff and Ready. 1991). Previous reviews have included this morphological description (Tomlinson, 1988; Ready. 1989; Cagan and Zipursky, 1992) and the molecules known to play important roles in retinal development (c.g. Rubin, 1989; Banerjee and Zipursky. 1991).

One of the most attractive aspects of the fly retina as a model system is the simplicity of its structure. The retina is composed of approximately 750 identical photoreceptive units, or ‘ommatidia’, arranged in a precise hexagonal array (Fig. 1). Each ommatidium is composed of exactly twenty cells. At the core of each ommatidium are eight photoreceptor neurons and four non-neuronal cone cells. These are surrounded by a lattice of pigment cells and bristles. The identity of each individual cell is defined by its position within the ommatidium, by its morphology, and by its expression of cell type-specific molecular markers. Thus, studies of cell fate in the retina can focus on the development of individual, identifiable cell types.

Recently, several laboratories have begun to identify some of the molecular mechanisms responsible for directing cell fate choices. Most of the protein components identified thus far appear also to play important developmental roles in other animals. These proteins include transcription factors, a steroid receptor, a receptor tyrosine kinase, and components of a tyrosine kinase signal transduction cascade. One of the most exciting outcomes of these studies has been to provide both a mechanistic and molecular link between seemingly disparate developmental processes studied in a variety of species. As a result, we are beginning to appreciate more fully how cells influence the cell fate choices of their neighbors.

Development of the fly retina begins in the syncytial blastoderm stage when approximately six cells arc set aside as retinal progenitors (Wieschaus and Gehring, 1976). During the next 4 days these cells proliferate into a monolayer epithelium known as the eye disc, but show no overt signs of differentiation until near the end of larval life. Cells cease dividing and begin differentiating in synchrony, forming a groove in the epithelium. This groove, known as the morphogenetic furrow, is the result of the coordinated basal movement of the nuclei of cells in the earliest stages of ommatidial differentiation. Differentiation, and hence the morphogenetic furrow’, begins first at the posterior edge of the eye disc and progresses anteriorly. Because ommatidia begin their differentiation at or near the morphogenetic furrow, the furrow provides a convenient ‘time zero’ for ommatidial maturity: ommatidia forming near the furrow are developmentally less mature than those further away (Fig. 2). As a result, each eye disc preparation contains a continuous temporal gradient of developmental stages. This gradient provides a tremendous advantage for studies of ommatidial development ami the mutations that perturb it.

The youngest ommatidial prcclusters typically contain six to seven cells. From this initial group, the first live cell types emerge: photoreceptor neurons R2. R3, R4. R5, and RS. Then, after the remaining undifferentiated cells traverse the cell cycle one last lime. RI. R6. and finally R7 are added to each ommatidial cluster to form the eight cell neuronal core (Fig. 2). The last cells to be added to the growing ommatidial core during larval life arc the four cone cells, which surround the inner eight neurons. The remaining cells are added during pupation, ami include the pigment cells and bristle organillos. During its development, each cell makes a series of dynamic, stereotyped cell contacts with its neighbors which may provide important information to the developing precursor cell (Tomlinson. 1985; Cagan and Ready. 1989b).

Al least two basic questions have been raised regarding larval ommatidial development. The first involves epithelial patterning: how do well-patterned ommatidial units emerge within an apparently homogeneous epithelial sheet? Little is understood about this process. Perhaps the only clue has come from studies involving DER. which encodes the Drosophila homologue of the epidermal growth factor receptor. A hyperactive allele of DER dramatically alters the number and spacing of ommatidia w ithin the eye disc (Baker et al., 1990). However, the nature of DER’s role in normal flies is as yet poorly understood. The second question regarding ommatidial development involves cell type differentiation. How arc cells within developing ommatidia directed to their proper fate, and how is this process of cell late selection initiated? This review will locus on the mechanisms of cell late selection within each ommatidium.

In the adult fly retina, the eight photoreceptor neurons can be divided into three groups based on their phototransduction machinery: RI-6. R7. and R8. During retinal development. these same cells can be further subdivided into five groups based on the symmetry of their position, the contacts each makes within the ommatidium, and the developmental time in which each group emerges (Fig. 3; see also Ready. 1989). Group one consists of R8. w hich sits initially at the ommatidium’s center. Group two neurons R2 and R5 Hank R8. Group three neurons R3 and R4 and Group four neurons R1 and R6 sit at opposite corners of the ommatidium. contacting groups one and two. And Group Jive is represented by R7. which sits between Rl and R6. If contacts between cells play a key role in directing cell fate (see below’), then the differing symmetry of the cells’ positions may provide opportunities for creating cell fate differences. Defining these live groups has proven prescient in at least one regard: the members of each group share a requirement for at least one gene activity critical in directing their cell fate choice. The significance of these live symmetry groups in specifying fate can be more fully tested when a greater umber of critical genes have been identified. The remainder of this review will discuss the role of the fewgenes already shown to play an important role in photoreceptor cell fate choice.

Within each developing ommatidium, the R8 neuron is the first cell type to extend an axon ami the first to express the neural-specific antigens 2200, 24BI0, and URP (Tomlinson. 1985: Tomlinson and Ready. 1987). By these criteria. R8 is the first terminally differentiating cell type in the developing retina. By what mechanism is the first cell type specified? Preliminary evidence suggests RS may be specified by at least a two step process (Fig. 4). First, large well-spaced pattern elements emerge within or just ahead of the morphogenetic furrow. These pattern elements have been visualized with a histological stain (Wolff ami Ready.1991) and by antibodies directed to the Scabrous protein (Baker et al., 1992). Each pattern element initially contains more than 20 cells, dwindling to a 6–7 cell pre-ommatidial cluster after emergence from the morphogenetic furrow. When Notch activity in the eye disc is reduced, most of the cells in the region of the morphogenetic furrow shift immediately into the neuronal pathway (Cagan and Ready. 1989a) and show characteristics of RS development (Baker et al., 1992). From this data it is templing to speculate that many of the cells within these large pattern elements are initially competent to differentiate as R8 neurons, and that ‘R8 competence’ is progressively narrowed.

After the 6–7 cell precluster is established, RS differentiation is further presaged by the emergence of a 2–3 cell group at its posterior edge. These cells slain more strongly with the histological stain cobalt sulfide (Cagan and Zipursky, 1992; Fig. 2 arrows), and exhibit distinctly stronger expression of a reporter gene under the influence of scabrous enhancer elements (unpublished results). The R8 neuron invariably derives from one of these 2–3 posterior cells. In some mutant backgrounds such as scabrous and rough, multiple cells within the group are often found to develop as RS neurons, resulting in 2–3 RSs within a single ommatidium (Fig. 5). Thus, both histological and genetic evidence suggest these cells form an ‘R8 equivalence group’. One simple model (Fig. 4) is that these cells are initially equivalent in their developmental potential, and that interactions between the cells select one to enter the R8 cell fate pathway. This model remains speculative, but morphological and genetic evidence suggests that a similar mechanism may be at work during other ommatidial cell fate decisions (Cagan and Ready, 1989a,b). For example, each member of the pigment cell lattice is typically derived from two or three candidate cells. Each of these apparent pigment cell precursors makes similar contacts, and the eventual ‘winner’ cannot be predicted before its differentiation begins.

Is there precedence for such a mechanism? In several respects, early ommatidial patterning resembles patterning of sensory organs described elsewhere in the fly. On the wing notum, for example, a large group of cells, distinguished by expression of members of the achaete-scute complex, forms the precursor pattern element to notum bristles: similar groups are seen throughout the developing nervous system (Campuzano and Modelell, 1992). Individual mother bristle cells emerge within this larger group to form bristle organule precursors. Surprisingly, despite the similarity of ommatidial and bristle development, achaeie and seule are not expressed or required in the developing retina. Furthermore, the presence of a 2–3 cell R8 equivalence group in the ommatidium has no described correlate in bristle development. A perhaps more similar example has been described in ablation and genetic studies involving vulval precursor cells (VPCs) of the developing nematode C. elegans (Sternberg, 1988; Sternberg and Horvitz, 1989). Interactions between three neighboring VPCs are required to give rise to a single primary vulval cell surrounded by two secondary vulval cells; blocking these interactions results in all three cells developing along the primary vulval cell pathway. Perhaps a similar competition is occurring between R8 precursors in the fly retina, leading to the production of a single R8 cell.

R2 and R5 emerge early in development as a symmetric pair of neurons which flank the R8 neuron. Other than their position within the ommatidium, these two cells exhibit no currently detectable molecular or morphological distinction between them. They express three neural-specific antigens - 22C10, 24B10 and HRP - just after the R8 neuron, suggesting their differentiation begins soon after R8 (this order is complicated by the fact that R2 and R5 express the neural antigen Elav just before R8). The rough gene product, initially expressed by most cells within the morphogenetic furrow, becomes restricted to R2 and R5 early in their development (Tomlinson et al., 1988). This expression appears to be important, as several studies indicate that rough plays a pivotal role in specifying the R2/R5 cell fate.

The rough locus encodes a nuclear protein that contains a homeodomain, suggesting that it acts as a transcription factor (Tomlinson et al., 1988). Genetic mosaic experiments indicate that only R2 and R5 require rough activity for an ommatidium to develop normally. In addition, mutations that eliminate rough activity prevent the R2 and R5 precursor cells (defined as the cells flanking R8) from achieving their normal fate (Heberlein et al., 1991; Van Vactor et al., 1991). Loss of rough activity also dramatically affects ommatidial organization; this is apparently a secondary consequence of the defects in R2 and R5.

Perhaps the strongest evidence that the Rough protein directs R2/R5 fate comes from ectopic expression experiments. By fusing a rough cDNA to a sevenless enhancer element, Rough protein can be expressed in several new cells, including the R7 precursor cell (“R7p”; Basler et al., 1990; Kimmel et al., 1990). The result is a transformation of R7p into a neuron whose adult morphology is characteristic of R1-R6. Thus expression of rough is sufficient to direct a cell into the R1-R6 fate, and perhaps the R2/R5 fate. We currently do not understand the importance of widespread rough expression in the morphogenetic furrow, nor how this expression becomes rapidly restricted to R2 and R5. Interestingly, loss of rough function can often redirect R2/R5 precursors into the R8 pathway (see above). Thus, the presence of rough can override both the R7 and the R8 developmental pathways. This suggests that a hierarchy of molecular pathways exists, and each cell may respond to the ‘strongest’ pathway to which it has access. How each cell’s developmental context directs gene expression, and hence its potential choices, remains a preeminent question. To this point, one clue may lie in the requirement for rough exclusively in the symmetric pair R2 and R5, two cells which also share a currently indistinguishable developmental history.

Soon aller R8, R2. and R5 emerge, photoreceptor neurons R3 and R4 show similar signs of differentiation, followed by Rl and R6. These four cells. R1. R3. R4. and R6 are not identical: R1 and R6 arise later than R3 and R4. and R4 eventually displays morphological movements which distinguish it from (he others. In addition, R1 and R6 express the Bar protein (Higashijima et al., 1992: sec below). However, these four cells, along with providing a symmetric group around R8. R2. and R5. have an important link. These four cells alone express and require the sevenup gene (Mlodzik et al., 1990). The sevenup locus encodes a protein with 75% identity to the human transcription factor COUP, a member of the steroid receptor superfamily. In the developing fly eye. sevenup activity is required for the correct specification of the RI/R3/R4/R6 fate.

Although the embryo requires sevenup activity for viability, the role of sevenup in the eye can be examined by generating mutant sevenup retinal cells through the use of mitotic recombination. These genetic mosaic studies indicate that sevenup activity must be present specifically in R1. R3. R4. and R6 for normal ommatidial development. In its absence, adult ommatidia contain three or more R7 neurons, apparently the result of a cell fate switch by the cells that require sevenup activity. As a putative transcription factor, it is templing to speculate that sevenup actively promotes the R1/R3/R4/R6 fate. Given its restricted expression, however, the precursors to these cells must already contain some cell type-specific information. The placement of sevenup in the steroid receptor superfamily suggests a possible role for diffusible factors in ommatidial development; however, a ligand for sevenup has yet to be identified.

Although Rl. R3, R4. and R6 make similar contacts within e developing ommatidium. Rl and R6 show several features that distinguish them. The development of Rl and R6 is temporally distinct: they are added pairwise from a round of cell divisions occurring after emergence of R3 and R4 The recent characterization of the Bar locus has added genetic and molecular distinction to Rl and R6 as well Higashijma et al., 1992). Deletions of the Bar locus result in a dramatic reduction in the size of the eye as well as a varicty of ommatidial abnormalities, and genetic mosaic experiments indicate that Rl and R6 require normal Bar activity for proper ommatidial development. The Bar locus encodes a pair of very similar homcobox-containing nuclear proteins. BarH I and BarH2. Consistent with their genetic requirement, both Bar transcripts are expressed in Rl and R6 as well as one of the later developing pigment cells (Higashijma et al., 1992).

While another gene that encodes a homeobox, rough. clearly has a role in cell fate specification, the role of the Bar locus is less clear. Most of the defects described have involved pigment cell development and secondary retinal tructures such as lens formation. In addition, the anterior of the eye is apparently more sensitive to al least some liar alleles (Sturtevant. 1925). Therefore, while Rl and R6 require normal liar activity, its role in their own development is not clear.

The best understood cell fate choice is that of photoreceptor neuron R7. Interest in R7 differentiation was triggered some twenty years ago by the discovery of a mutation in the sevenless locus, which results in loss of the R7 neuron (Harris et al., 1976; l ig. I inset). Tomlinson and Ready (1986) demonstrated that loss of R7 in sevenless ommatidia was a result of the R7 precursor cell (R7p; defined by its position within the ommatidium) developing incorrectly as a non-neuronal cone cell. The sevenless locus encodes a transmembrane protein with a large extracellular domain and a cytoplasmic consensus tyrosine kinase catalytic domain (Hafen et al., 1987). The discovery that sevenless encodes a receptor tyrosine kinase suggested a simple model for sevenless activation of the R7 pathway. Presumably an extracellular signal activates the sevenless receptor, which in turn activates a signal transduction pathway to bring this signal from the surface to the nucleus, directing R7p into the R7 pathway.

The best candidate to provide a ligand for Sevenless was the gene product encoded by bride of sevenless (boss). Mutations in boss give rise to the same phenotype as those in sevenless: R7p develops as a cone cell (Reinke and Zipursky. 1988; Cagan et al. 1993). In addition, genetic mosaic experiments indicated that boss activity is required exclusively in the neighboring R8 neuron. Could RS provide the inductive ligand? The sequence of boss (Hart et al., 1990) suggests a surprising structure for a possible ligand: hydrophobicity plots indicate Boss is a transmembrane protein with seven membrane-spanning domains. Although boss’s sequence indicates no significant amino acid identity to any protein currently reported, its overall structure would seem to place Boss within the family of seven transmembrane-containing G protein-coupled receptors. Despite its resemblance to a receptor, both tissue culture and in vivo studies indicate Boss is a membrane bound ligand of the Sevenless receptor.

Tissue culture studies provided evidence that Boss can bind directly to Sevcnlcss (Kramer et al., 1991). When bossexpressing cells were mixed with sevenless-expressing cells, large aggregates formed composed of both cell types. A series of control experiments demonstrated that this aggregation was due specifically to the direct binding of Boss and Sevcnlcss. Furthermore, binding was shown to result in phosphorylation of the Sevcnlcss receptor (Hart et al., 1993). a common first step in activation of receptor tyrosine kinases. Whether this represents autophosphorylation and whether the phosphorylated form of the Sevenless receptor represents its active stale has yet Io be established.

To determine whether Boss is a ligand of (he Sevenlcss receptor in vivo. Krämer et al. (1991) examined its distribution in the developing eye (Fig. 6). Boss protein was localized primarily to the apical microvilli of the developing R8 photoreceptor neuron. In addition. Boss immunore activity was detected in the perinuclear endoplasmic reticulum of R8 alone, indicating that Boss is made exclusively by R8. Nevertheless. Boss immunoreactivity was also delected within a large vesicle in the developing R7p. This vesicle, referred to as a ‘multivcsicular body’, is structurally similar to late endosomes, compartments within which internalized ligand/reccplor complexes arc typically sorted (Trowbridge. 1991). Thus. Boss protein is internalized into R7p from the neighboring RS neuron. This internalization requires presence of the Sevenless receptor (Fig. 6): mutations that eliminate the Sevenlcss protein prevented Boss internalization into R7. while a mutation that inactivates the Sevenlcss protein but allows normal expression permitted Boss internalization (Kramer et al., 1991). In other words. Boss internalization requires only physical presence of the Sevenless protein, not its activity. Together with the tissue culture studies, these experiments provide strong evidence that Boss is indeed a (rather unusual) ligand of the Sevenless receptor in the developing retina.

Surprisingly, experiments involving antibodies directed to both extracellular- and cytoplasmic-localized epitopes indicated that the entire Boss protein finds its way into R7’s multivesicular body (Cagan et al., 1992). The mechanism by which a seven-membrane spanning protein is removed from one cell and internalized into its neighbor is currently unknown. Interestingly, inactivating the Shibirc protein, required for receptor-mediated endocytosis (Kosaka and Ikeda. 1983). blocks Boss internalization into R7. suggesting a role for the endocytotic pathway in the internalization Boss (Cagan et al. 1992). The importance for a cell to internalize Boss is not clear. Truncated, constitutively active forms of the Sevenless receptor can direct R7 development even in the absence of Boss, suggesting that activation of Sevenless alone is sufficient to trigger R7 development Basler et al., 1991). However. Boss could still play a role keeping the Sevenless receptor active within R7p. In addition, the truncated Sevenless receptor was expressed at. very high levels: this expression could conceivably override any requirement for internalized Boss. Perhaps the unusual structure of Boss reflects its additional role as a receptor. However, development of the R8 neuron appears unaffected by the loss of Boss, so its potential for receptor-like activity is unclear.

Thus, studies involving R7 cell fate selection have provided us with one of the first examples of a local inductive event in which the molecular nature of the information passed between cells has been identified. By presenting Boss to R7p. RS provides a membrane-anchored inductive cue which binds to and activates the Sevenless receptor tyrosine kinase. Activation of Sevenless is sufficient to activate the R7 developmental pathway in competent cells (Basler et al., 1991). suggesting the interaction of Boss and Sevenless may represent the sole inductive interaction between R7 and RS that directs cell fate. This interaction raises several interesting developmental questions.

One important question involves the nature of the restriction of the boss inductive signal. Why docs each ommatidium contain only a single R7 cell? This restriction is not likely ‘ be due to the receptor, as sevenless is expressed in most .ells in the developing eye disc (Tomlinson et al., 1987: Banerjee et al., 1987). Indeed, expressing sevenless in all cells still results in a single R7 cell in each ommatidium (Basler and Hafen, 1989; Bowtell et al., 1989). Studies with boss indicate that one important restriction is the specific tethering of Boss to R8. Ubiquitous boss expression, achieved by fusing boss cDNA to an inducible promoter, results in transformation of developing cone cells into R7 neurons (Van Vactor et al., 1991). Cone cells do not contact R8 early in their development (Fig. 3) and therefore arc not ormally in contact with the Boss inductive ligand. Apparently cone cells arc indeed competent to respond to the Boss inductive cue. and when ectopic expression presents Boss to icm they will respond by activating the R7 pathway.

By contrast, photoreceptor neurons R1-R6 are in direct contact with (boss-exprcssing) R8. yet none activate the boss/sevenless pathway. This is surprising since, for example. R3 and R4 express sevenless at levels similar to R7. Yet not only do they not activate the boss/sevenless pathway, they also fail to internalize Boss into their own multivesicular bodies. When the activity of genes involved in specifying RI-R6 - such as rough and Notch - is disturbed, many of these cells will then internalize Boss and develop as R7 cells. This result suggests that the R7 fate may represent a default state, and that proteins such as those encoded by rough and Notch may be required to redirect cells into a different fate. One result of this redirection is the establishment of a block to Boss internalization.

Thus, the ommatidium can be divided into at least four groups with respect to the boss/sevenless pathway. The first group is represented by the R8 neuron, which provides the inductive cue to its immediate neighbors. The second group is represented by neurons R1-R6. which are neither competent to respond to the Boss inductive cue, nor will they internalize Boss. This is due to their previous activation of alternative developmental pathways through genes such as rough and Notch. The third group is represented by R7p. which responds to the inductive cue by activating the sevenless pathway and internalizing Boss. And the final group is represented by the four cone cells, which arc competent to respond to the inductive cue but are normally a full cell diameter away from Boss; by this mechanism they arc freed to develop along the cone cell pathway.

Rapid progress has been made recently in understanding the nature of the molecular pathway triggered by an activated Sevcnless receptor (Fig. 7: sec also Halen, this volume, for a more extensive review). Screens for genes that interact genetically with sevcnless have indicated that members of the rus signal transduction pathway play a key role in bringing the sevcnless signal to the nucleus (Simon et al.. 1991). Drusl encodes the Drosophila homologue of p2l^ras. a GTPase implicated as a downstream component of most receptor tyrosine kinase pathways examined (Cantley et al., 1991). Mutations in Drasl interact genetically with sevcnless mutations and prevent normal photoreceptor development, including R7. Strikingly, when a constitutively active form of Drasl is expressed in the eye. it can direct cells into the R7 pathway even in the absence of sevcnless activity (Fortini et al., 1992). This strongly implicates Drasl as a downstream component of the sevcnless pathway.

Three genes which are good candidates to act upstream of Drasl in the developing eye are GAP1, Sos. and drk. Gap1 encodes the Drosophila homologue of a family of proteins known as GTPase activating proteins, or GAPs (Gaul et al., 1992). GAPs have been implicated as direct inactivators of p21^ras due to their ability to catalyze its intrinsic hydrolytic activity (Bourne et al., 1991). Conversion of p2l^ras-GTP to p2l^ias-GDP inactivates its GTPase activity. Consistent with a role for GAP1as a direct negative regulator of Drasl. mutations that reduce GAP1 Activity increase the number of cells which develop as R7 cells. GAP! mutations can occasionally override R7p’s requirement for scvenlcss activity (Gaul et al., 1992), but most still require scvenlcss to develop as R7s (Rogge et al., 1992). Thus, other genes must also regulate the propagation of sevenless activity.

Mutations that inactivate Sos or drk activity have the opposite phenotype to GAP1: photoreceptor cells arc lost, with the R7 cell being the most sensitive to loss of either activity (Simon et al., 1991; Bon fini et al., 1991; Simon et al., 1993; Olivier et al., 1993). Sos encodes the Drosophila homologue of a family of guanine nucleotide exchange factors (GNEFs) which activate the ras pathway by catalyzing an increase in active GTP-bound forms of p21^ras (Brock et al., 1987: Jones et al., 1991). The drk gene encodes a small adaptor protein consisting almost entirely of an SH2 domain surrounded by two SH3 domains. Biochemical studies indicate that drk can bind directly to both sevenless and Sos. This suggests a simple model in which sevenless activates Sos through drk. and Sos in turn activates Drasl (Simon et al.. 1993; Olivier et al., 1993).

More recently, several genes have emerged as candidates to lie downstream of Dras1 in the sevenless signal transduction pathway. One is Draf. which encodes the Drosophila homologue of mammalian ral kinase. Mutations that eliminate activity in Draf block cell proliferation and are lethal, while weak, viable Draf alleles affect photoreceptor differentiation, especially R7 (Dickson et al., 1992). Similarly to Drasl. artificially high levels of/hq/’can direct development of ectopic R7 neurons. When one copy of Draf is removed in a fly which contains activated Drasl (see above), fewer ectopic R7s are recruited, suggesting that Drasl acts through Draf (Dickson et al., 1992).

Two other candidates for acting in the sevenlcss pathway are encoded by the Dsor and DmERKA loci. Both are cytoplasmic kinases. Dsor encodes the Drosophila homologue of MAP kinase kinase or MEK (Tsuda et al., 1993). Genetic evidence places Dsor in the torso signal transduction pathway. The torso gene encodes a receptor tyrosine kinase (Sprenger et al.. 1989). and it activates a signal transduction pathway which contains many of the same components as the sevenlcss pathway. Dsor is required for embryo viability, and its role in retinal development in larva has yet to be established. Due to its homology to MAP kinase kinases. Dsor is a good candidate to provide a direct link between Raf and DmERKA. the Drosophila homologue of MAP kinase (Biggs and Zipursky. 1992). DmERKA is expressed in the developing eye disc. Whether DmERKA plays a role in eye development must await the isolation of DmERKA mutants.

Based on morphological studies in the ant (Bernhard. 1937). lineage was thought to be the sole determining factor for cell fate selection in the developing arthropod retina. This view has changed, as genetic mosaic studies (Ready et al.. 1976; Lawrence and Green. 1979) indicated that a cell’s lineage does not direct its choice of fate in the fly retina (the role of lineage in the ant has not been further examined). At least four lines of evidence support the view that cell fate choice requires interactions among neighboring cells. The first evidence, discussed above, is the stereotyped contacts made between a developing precursor cell and its neighbors. Morphological and histological studies (Tomlinson. 1985; Tomlinson and Ready. 1987; Cagan and Ready, 1989b) have led to the suggestion that cells that have begun their own differentiation provide inductive cues to their uncommitted neighbors through direct physical contact.

The second line of evidence for the importance of local interactions emerged from studies involving the rough locus. Normal rough activity is required exclusively within R2 and R5 for proper ommatidial assembly (Tomlinson et al., 1988; Hcbcrlein et al.. 1991; Van Vactor et al., 1991; see above). Nevertheless, rough mutations also affect R3 and R4 development (Tomlinson et al., 1988). leading to the suggestion that R2 and R5 provide local information required for the correct development of R3 and R4. This interpretation is complicated by the observation that elimination of rough activity also dramatically alters the positions of other cells within the ommatidium (Van Vactor et al., 1991).

Studies involving the role of Notch in the retina provide further evidence for the importance of local interactions. Notch encodes a large transmembrane protein (Wharton et al., 1985; Kidd et al., 1986) which is required cell autonomously in both the retina (Cagan and Ready, unpublished) and elsewhere in the fly (Hoppe and Greenspan. 1990; Heitzler and Simpson, 1991). Notch has been implicated in mosaic studies as an important mediator of communication between neighboring cells (Heitzler and Simpson. 1991). Using the temperature-sensitive allele N^ts1. Notch activity was reduced for brief periods throughout retinal development (Cagan and Ready, 1989a). The fate of each cell type was altered when Notch activity was reduced early in its development. In addition, switching the fate of one cell (e.g. a cone cell) often changed the fate of its later-differentiating neighbors (e.g. neighboring pigment cells). These results were interpreted as evidence that cells require local interactions during the time of their cell fate selection, and that these interactions make use of Notch activity.

The strongest evidence of a role for cell-cell interactions in directing cell fate is provided by the well-characterized interaction between R8 and R7. described above. The direct binding of transmembrane proteins Boss and Sevenless in vitro and Boss internalization in vivo provide clear evidence that this interaction is mediated by physical contact between the two cells. However, we do not yet understand the precise role of the sevcnlcss pathway during differentiation of R7p.

Expanding rough expression can direct R7p into the RI-R6 developmental pathway (see above), indicating rough directs cells into a particular fate. Does sevenless have a similar role in directing the R7 fate? This is an especially important question, because the sevenless pathway is the only one clearly shown to be activated by a neighboring cell. In other words, do cell-cell interactions specify fate or do they simply direct a cell to act on its own intrinsic information. Ectopic expression of an artificially activated Sevenlcss receptor can direct cone cell precursors into the R7 pathway, but fails to alter photoreceptor development (Dickson et al., 1993). The fact that different cells respond differently to activated Sevenless suggests that cells contain intrinsic donnation which shapes their responses to external cues.The nature of this intrinsic information may be hinted at in re experiment involving expanded expression of rough. As described above, when R7p expresses rough it develops as an R1-R6 cell. Remarkably, this transformation still requires sevenless activity; in its absence R7p still developed as a cone cell even in the presence of rough expression (Basler et al., 1990; Kimmel et al., 1990). Perhaps the simplest explanation is that the activated Sevcnlcss receptor provides. necessary cue. but a cell s own intrinsic information iccidcs the manner in which it responds. This intrinsic information has already limited R7p’s developmental potential to o possible choices: R7 or cone cell fate. By adding rough .dix ity, this choice is increased to three. In other words, the information imparted by sevenless alone may not be enough to determine cell fate in this context.

The early morphological movements of R7p also suggest its differentiation is in part independent of sevenless function. The R7 neuron begins its differentiation (i.e. expression of neuronal markers) between R1 and R6 approximately a day after ommatidial development begins, and this differentiation requires sevenlcss activity. Rl and R6 begin their own differentiation several hours earlier than R7. and they are invariably separated by a single cell. This cell is R7p, and it is positioned between Rl and R6 even in the absence of boss or sevenless (Fig 2; unpublished data). In other words. R7p has an extensive developmental history independent of the sevenless pathway. This developmental history is likely to provide the information necessary to limit R7p’s developmental potential. Perhaps it is this information with which sevcnlcss interacts.

As the sevenless pathway has become belter understood, the importance of cell-cell communication has been confirmed, implicating a commonly used signal transduction pathway as a mediator of cell signaling. In addition, studies involving cell fate specification of other photoreceptor cells have implicated several types of transcription factors as important elements in determining a cell’s fate. As the molecular aspects of ommatidial development become belter understood, we will be in a belter position to link these factors to the more general pathways represented by sevenless. In addition, we will be belter able to understand the influence that a cell’s position within the retinal epithelium has on its developmental potential: the link between morphology and molecules.

Thanks to Dr Larry Zipursky. Dr Francesca Pignoni. and Dr Tony Brown for helpful comments. The author is supported by a grant from the American Cancer Society.